Discrimination on 256 shape-color combinations (test)
99.5Test AccuracyDINOv2
Evaluation Results
| Method | Links | |
|---|---|---|
| DINOv2Backbone=ViT-B/16, Pretraining Scale=142M2024.06 | 99.5 | |
| CLIPBackbone=ViT-B/16, Pretraining Scale=400M2024.06 | 99.3 | |
| MAEBackbone=ViT-B/16, Pretraining Scale=1.28M2024.06 | 98 | |
| ImageNetBackbone=ViT-B/16, Pretraining Scale=14.2M2024.06 | 97.5 | |
| DINOBackbone=ViT-B/16, Pretraining Scale=1.28M2024.06 | 95.6 | |
| ViT-B/16 (No pre-training)Backbone=ViT-B/16, Pretraining Scale=None2024.06 | 80.5 |